an experience i thought i would share. For the past few days i have been converting my entire cd collection to mp3's and uploading them to my empeg. the other day i decided to plug it into my stereo and listen for a bit. i currently have over 1200 songs uploaded and when i shuffled the player at one point these three songs played one after another

Nirvana - Polly - Unplugged in New York
Nirvana - Polly - Nevermind
Nirvana - Polly - From the muddy banks of the Wikasha

with these unlikely odds seems i should go buy a lottery ticket.

In reply to:

---

with these unlikely odds seems i should go buy a lottery ticket.

---

That's pretty good. That's a 1 / 287,280,400 chance by my calculations. Lotteries typically run about 1 / 26,000,000.

I am just short of 6000 songs and I had three NIN tracks come up in a row yesterday afternoon. And within 10 tracks in either direction, 2 more NIN tracks.

The odds would be astronomical, if they empeg's shuffle wasn't somehow buggered and responsible for this.

And if you want to figure out odds, you should look at the odds for 3 songs by any given artist, not for three specific tracks (because those odds are the same for any other 3 specific tracks). I have a few NIN albums which contain about 78 songs and then some others from compilation discs (let's say 85 total to be very safe).

Bruno

In reply to:

---

And if you want to figure out odds, you should look at the odds for 3 songs by any given artist

---

Well, in your case, the odds of any three NIN songs coming up at the beginning of a new shuffle is a mere 1 / 364,000. The calculations get more complicated when you start talking about three songs by NIN showing up in a row anywhere in the middle of a playlist, but they drop dramatically, to somewhere around 1 / 60 (not entirely sure of that number, but it should be close).

In reply to:

---

not for three specific tracks

---

But in his case it was three specific tracks. I am assuming that those are the only three versions of "Polly" that he has on the Empeg. The order doesn't matter, and I accounted for that, but it had to be specifically those three songs and no other Nirvana ones.

An interesting statistic. Assuming you have 85 NIN songs, and you randomized just them, what are the odds that they would play in the correct order as if you hadn't shuffled them at all? 2.8 X 10^128 !

This is all assuming of course that it is truly random and the way the Empeg does it isn't, as Bruno said, "buggered".

If this type of thing bugs you, you can setup a playlist specifically designed to not the same play a song from the same artist back to back. (You set up a playlist for each artist under one massive playlist and set the option to only select one song from each of these artist playlists). There was a thread a couple of weeks ago about this, and it works great, as long as you don't mind taking a liitle bit of time and effort to set it up. Only problem for me is, every time I select my list I end up with at least one of the same songs. Even if there are 20-30 songs by an artist, I will often hear that same song again (only once through the list though, so you wont get the effect you had). Hopefully they will be putting in some way of preventing this in a future (post 2.0 I am expecting) release.

-Jeff

Crap, I only noticed that it played 3 copies of Polly after I posted. I've only ever had this phenomenon with two tracks. It would have been bad enough, as in my case, just playng the same artist 3 times in a row. Neat.

Finite math was a cool class in highscool. Haven't had much cause to use anything complex over the past few years though.

Bruno

If you were using the "Least Recently Played" version of the shuffle, and those three tracks were part of an album that had not been played recently (or had not ever been played because it had just been uploaded), then the shuffle is working as expected.

"That's a 1 / 287,280,400"

I believe it's 1/1,723,682,400. 1200 x 1199 x 1198. That is if you're talking about any 3 given consecutive songs being the 3 specific tracks.

And of course a computer can't pick a truly random number. It can only follow an algorithm. That makes you wonder whether there is such a thing as random.

> I believe it's 1/1,723,682,400. 1200 x 1199 x 1198.

Yeah, that's the odds if order mattered. I was going under the assumption that the order did not and the three songs could appear in any variation, as long as they were together.

Setting up the playlists as you specified would definitely work, but I'd much rather see the issue fixed in the empeg itself. I think we all agree that this happens a lot more often than to be considered "coincidence".

It's obvious the empeg "random" shuffle isn't perfectly random. I'd be curious to see what criteria/algorithm they use to pick the songs while in random shuffle mode.

It's obvious the empeg "random" shuffle isn't perfectly random

Yes, sadly we have no particle driven random number generator on board.

Rob

Yes, but my Rio Volt does a much better shuffle than the empeg, and it's not nearly as advanced.

Yes, sadly we have no particle driven random number generator on board.

But rumo[u]?r has it that's on the post-2.0 wish list. Of course if 2.0 never goes final you don't have to worry about post-2.0 features! Do I smell a conspiracy?

Heh, they could be like ICQ. I don't think that software will EVER get out of beta.

Actually it's only approximately 1 in 720,000. (Assuming there is only 3 versions of 'Polly').

We are playing the whole playlist, so at sometime we will find the first version. We don't care which version it is, or where in the playlist it is. The odds of it being there are 1 in 1.

The question is about which tracks follow. Now we are looking for 1 of 2 tracks out of 1200, without caring which of the 2 we find. Odds of 1 in 600. Finally we need one track out of 1200.

So 1/600 * 1/1200 = 1/720,000.

It's not worth bothering about adjusting for 1199, 1198 etc. He said 'over 1200' tracks, so worrying about this is pointless. He could have 120000 tracks for all we know, which makes our error extreme. Of course, the context suggests that he has just over 1200 tracks, but you get the point..

You know if you count all the empeg owners that have the same song repeated three times on their units (I have at least three versions of "Closer To The Heart"), and the number of times they have shuffled the contents of their player, and the odds being 720,000 that one of these shuffles would hit three songs in a row- its really not outside the realm of possibility that he could have just been the "lucky" one. I too have felt the player's "random" is not quite right - but I'm not sure of it. Keep in mind these three songs he was talking about were not all on the same album, and we don't even know if they were loaded close to one another. If not, this "broken randomness" would work against his odds of getting the three songs in a row. Of course, he probably loaded them all on at the same time so they are probably close, but still not on the same album. Just my observations

-Jeff

Well, if there are 4000 owners, then the odds of it happening to someone in any given playlist shuffle is:

1/720,000 * 4000 = 1/180

This is assuming that we all have 1200 tracks loaded, 3 of which are different versions of the same track.

How many days listening is 1200 tracks?

How many days listening is 1200 tracks?

I don't think anyone ever gets all the way through a playlist that large.

My player gets used constantly, and I rarely get more than a couple hundred tunes into a large shuffle before I've re-shuffled or selected another playlist for some other reason.

For instance, I just got back from a weekend trip with my wife. When she's around, the player gets dropped into its "low key" playlist and stays there. That's currently a list of about 250 songs.

We had the player going for the entire 2.5-hour trip to Napa, we had it hooked up in the hotel room playing about an hour or two each day, it played for about an hour as we drove around Napa, and we had it playing for the entire drive back.

It only got about 2/3 of the way through this relatively small playlist by the time I got home and put it back into a full-player-shuffle.

I've watched this problem for a while. What I've seem to have noticed is that the player, when picking songs randomly, sometimes will pick multiple songs out of the same playlist and play them back-to-back.

For example, say you have the following playlists:

Alternative Rock
Classic Rock
Country
Hip-Hop
Singles
Swing

For the sake of argument, lets also assume there are tons of other playlists so that we can eliminate the "but that's only 6 playlists".

A typical random shuffle might look something like this:

Country song
Alternative Rock song
Random Playlist1 song
Swing song
Random Playlist2 song
Classic Rock song
Hip-Hop song
Random Playlist3 song
Random Playlist4 song
Singles song
etc...

However, I've noticed that when I shuffle, I tend to get playlists more like this:

Country song
Alternative Rock song
Random Playlist1 song
Swing song
Swing song
Random Playlist2 song
Classic Rock song
Hip-Hop song
Random Playlist3 song
Hip-Hop song
Random Playlist4 song
Singles song

I tend to notice grouping of songs from the same playlist. Sometimes it's from the same CD...sometimes it's from the same genre playlist.

Take a look yourselves for a week or two and tell me you don't notice this more often that what could fall under "chance".

I have also noticed this "combing" effect and commented on it elsewhere on the BBS.

ninti,

True. Good catch.

svferris,

Say you have ten playlists. Say the first song in your root playlist (all playlists) is from alternative rock. The chances of the next song being from alternative rock are 1 in 10 (assuming you have same-sized playlist). So the chances of no songs from the same playlist playing back to back is very small.

Perhaps a future version could have an option other than random that blends the different generes/artists evenly, kinda like a radio station. You won't hear the same artist back to back on the radio, unless they do it on purpose.

To all,

How can you measure perfect randomness? It could pick the same order 100 times in a row, yet still be random.

> I don't think anyone ever gets all the way through a playlist that large.

Thank god for that new feature of bookmarks. I have a playlist consisting of my 2500 favorite songs. I have been listening to it for about a month or so and am only 10% through. When I feel the need to listen to something else I bookmark it and come back for it later.

Those Empeg guys think of everything.

I've had my share of freaky shuffles. My empeg usualy picks a few artist it likes every night and features them more than any other. Probably just my imagination...

The really wierd one was the day George Harrison died. Every other song seemed to be the beatles.

Say you have ten playlists. Say the first song in your root playlist (all playlists) is from alternative rock. The chances of the next song being from alternative rock are 1 in 10 (assuming you have same-sized playlist). So the chances of no songs from the same playlist playing back to back is very small.

Yes, but it my case, I have 2,200 songs and at least 75 playlists (probably a lot more). Now, I admit that some playlists are much bigger than others. So, let's take something like my Tenacious D playlist. I've seen tracks from this come back-to-back before (or within a few songs). There are about 20 tracks in the playlist. That means the chance of getting one of those tracks at any time would be 20/2200 or 1 in 110.

Now, the chance of one of those songs coming up as the next song is still approx. 1 in 110 because it's independent odds.

But (and correct me if I'm wrong), the probability of the two coming back to back would be 1/110*1/110, which is significantly smaller chance.

> Actually it's only approximately 1 in 720,000

Oops, yeah, you got me. I was going under the assumption that those were his FIRST three songs, which is, of course, not what he actually said. Apparently Bruno was not the only to mis-read the post. All that math knowledge won't do me any good if I can't solve the correct equation. And I always considered word problems to be one of my skills.

Hmm, but let me explore your solution a little bit more thoughly. Your example works IF the FIRST song is one of the Pollys. Then the chances of the next song being Polly would indeed be 1 / 600. But if the first Polly was played half way through, then the chance of the next one being Polly would only be 1 / 300.

To take a simpler example, say we have three songs; two Pollys and a Plateau. What are the odds that the Pollys would be together if shuffled? A simple counting of the alternatives would see that it is 4 / 6. If we followed your logic that the odds of the first one being found is 1 / 1, and then the odds of the second one are 1 / 2, we are left with the odds being 1 / 2.

On the other hand, if the first one is Polly ( a 2/3 chance), then there is a 1/2 chance of the second being polly. If the first one is not Polly ( a 1/3 chance), then the second and third will always be pollys (1/1). So 2/3*1/2+1/3*1/1 = 2/3 or 4/6. I checked it with four choices as well and that method seems to hold up. Obviously I can't do that for 1200 songs, though I am sure there is an easier way that I don't know about. It has been a while since I took statistics.

> Now, the chance of one of those songs coming up as the next song is still approx. 1 in 110 because it's independent odds.

But if you are already on a given song, then the chances are just 1 / 110 that the next song will be off the same playlist. What you are talking about applies if you are talking about specific positions. The odds of a Tenacious D song being the the first song (or song 666 or whatever) are 1 / 110, and the odds of another one being the next song after that are (about) 1 / 110 * 1 / 110.

In reply to:

---

> Actually it's only approximately 1 in 720,000

Oops, yeah, you got me. I was going under the assumption that those were his FIRST three songs, which is, of course, not what he actually said. Apparently Bruno was not the only to mis-read the post. All that math knowledge won't do me any good if I can't solve the correct equation. And I always considered word problems to be one of my skills.

Hmm, but let me explore your solution a little bit more thoughly. Your example works IF the FIRST song is one of the Pollys. Then the chances of the next song being Polly would indeed be 1 / 600. But if the first Polly was played half way through, then the chance of the next one being Polly would only be 1 / 300.

---

Good points. Let's look at this a different way. Say that there are n songs in the playlist, k of which are Polly. Then there are exactly C(n,k) = n!/(n-k)!k! ways to distribute the k Polly's among the n songs, and there are n-k+1 ways to place k consecutive Polly's in n songs. Plugging in n=1200 and k=3 gives us a probability of 1198/287280400 or 1/239800.

All assuming all shuffles are equally probable which we know isn't the case (due to Empeg bugginess and the limitations of psuedorandom generators). BTW, I've had two versions of the same Tori Amos song pop up in a row which surprised me quite a bit. But now that I've done the math it doesn't seem quite so unusual.

And, just as a counter example to Tony's claim, I've actually cycled through my 1400+ song playlist multiple times. I was doing a *lot* of driving last fall.

--John

Wow, it's been a while since I've discussed this much probability. Brings me back to my CS classes in college.

Although I still think the shuffle isn't as random as it could be, I do somewhat like it because the empeg tends to act more like a radio station. You know how they occasionally will do a double play from an artist, or they'll play blocks of genres, such as two female vocals followed by two hard rock, etc.

With that in mind, I thought I'd do a poll:

In reply to:

---

Good points. Let's look at this a different way....Plugging in n=1200 and k=3 gives us a probability of 1198/287280400 or 1/239800.

---

Yeah! That's it. Hats off to you man. I sure hope the original poster didn't buy his lottery ticket after all.

Okay, let n be the number of songs. k of which are polly. L is the number of shuffles, H is the number of songs going to be played of n. B is the number of H of n are k. So far we have:

k= [n x L / (H - B + k) x k^2] / radical n^2.

Follow? Now if we want the probality of n for L with respect to H as B with a quotient of k, we'll need to define a few more varibles. G is k / B, and Y is the number of songs played minus songs that will play divided by the square of total songs. S is y + (k^2 x G). So,

H= S^2(Y^2 - cos(B-G) + k^2).

Now if we multiply that times the square root of the serial number we find that it is a 1 / 1 chance that k= BS.

Careful Yz33d, you're being pretty funny again. You wouldn't want this to damage your reputation would you?

Well, if you seed the generator using the microphone input?

Calvin

In reply to:

---

I don't think anyone ever gets all the way through a playlist that large.

---

I just finished a 2765-track playlist, about 8d 13 h of playing time or six weeks of listening (because I listen to radio in the car, and small playlists when going to sleep). Bookmarks are the dog's back wheels for this job.

And yes, I did listen to it all - it's my "all albums" playlist, with the shuffle flag set, and I was listening for inter-track gaps (and populating seek tool data in the process).